Spatiotemporal neuromorphic clusters improve event sensor data handling
A Spatiotemporal Extension of the Neuromorphic DBSCAN Implementation
Neural and Evolutionary Computing
Summary
Clustering data is a way to find groups and remove noisy information. The authors had already created two different hardware designs to run a clustering algorithm called DBSCAN more efficiently using brain-inspired computer chips. Now, they extended both designs to better handle data that changes over time and space, like from special sensors that detect events continuously. Their updated designs can also be adjusted to use less hardware by trading off speed, making them more practical for real devices. The authors shared all their designs openly for others to use.
What this means in practice
- •For embedded systems engineers: Build energy-efficient, event-driven clustering modules for real-time sensor data processing in hardware-constrained devices.
- •For robotics developers: Integrate spatiotemporal clustering to improve noise filtering and pattern recognition from event-based sensors in robotic perception systems.
Authors
Charles P. Rizzo, James S. Plank
Abstract
DBSCAN is an algorithm that denoises and clusters data. In prior work, we implemented the DBSCAN algorithm neuromorphically, introducing two constructions termed ``flat'' and ``systolic''. The ``flat'' construction prioritizes throughput, while the ``systolic'' construction trades time for space resulting in a smaller, more hardware-friendly architecture at the cost of throughput. In this work, we offer spatiotemporal extensions of these two constructions to better leverage the spatiotemporal nature of event sensor data. Moreover, as in our prior work, we discuss partial or segmented implementations that further leverage time for space when hardware resources are constrained. All network constructions are provided as open-source implementations.